projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
perf report: Fix placement of itrace option in documentation
[deliverable/linux.git] / tools / perf / util / session.c
1 #include <linux/kernel.h>
2 #include <traceevent/event-parse.h>
3
4 #include <byteswap.h>
5 #include <unistd.h>
6 #include <sys/types.h>
7 #include <sys/mman.h>
8
9 #include "evlist.h"
10 #include "evsel.h"
11 #include "session.h"
12 #include "tool.h"
13 #include "sort.h"
14 #include "util.h"
15 #include "cpumap.h"
16 #include "perf_regs.h"
17 #include "asm/bug.h"
18 #include "auxtrace.h"
19
20 static int perf_session__deliver_event(struct perf_session *session,
21 union perf_event *event,
22 struct perf_sample *sample,
23 struct perf_tool *tool,
24 u64 file_offset);
25
26 static int perf_session__open(struct perf_session *session)
27 {
28 struct perf_data_file *file = session->file;
29
30 if (perf_session__read_header(session) < 0) {
31 pr_err("incompatible file format (rerun with -v to learn more)");
32 return -1;
33 }
34
35 if (perf_data_file__is_pipe(file))
36 return 0;
37
38 if (!perf_evlist__valid_sample_type(session->evlist)) {
39 pr_err("non matching sample_type");
40 return -1;
41 }
42
43 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
44 pr_err("non matching sample_id_all");
45 return -1;
46 }
47
48 if (!perf_evlist__valid_read_format(session->evlist)) {
49 pr_err("non matching read_format");
50 return -1;
51 }
52
53 return 0;
54 }
55
56 void perf_session__set_id_hdr_size(struct perf_session *session)
57 {
58 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
59
60 machines__set_id_hdr_size(&session->machines, id_hdr_size);
61 }
62
63 int perf_session__create_kernel_maps(struct perf_session *session)
64 {
65 int ret = machine__create_kernel_maps(&session->machines.host);
66
67 if (ret >= 0)
68 ret = machines__create_guest_kernel_maps(&session->machines);
69 return ret;
70 }
71
72 static void perf_session__destroy_kernel_maps(struct perf_session *session)
73 {
74 machines__destroy_kernel_maps(&session->machines);
75 }
76
77 static bool perf_session__has_comm_exec(struct perf_session *session)
78 {
79 struct perf_evsel *evsel;
80
81 evlist__for_each(session->evlist, evsel) {
82 if (evsel->attr.comm_exec)
83 return true;
84 }
85
86 return false;
87 }
88
89 static void perf_session__set_comm_exec(struct perf_session *session)
90 {
91 bool comm_exec = perf_session__has_comm_exec(session);
92
93 machines__set_comm_exec(&session->machines, comm_exec);
94 }
95
96 static int ordered_events__deliver_event(struct ordered_events *oe,
97 struct ordered_event *event)
98 {
99 struct perf_sample sample;
100 struct perf_session *session = container_of(oe, struct perf_session,
101 ordered_events);
102 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
103
104 if (ret) {
105 pr_err("Can't parse sample, err = %d\n", ret);
106 return ret;
107 }
108
109 return perf_session__deliver_event(session, event->event, &sample,
110 session->tool, event->file_offset);
111 }
112
113 struct perf_session *perf_session__new(struct perf_data_file *file,
114 bool repipe, struct perf_tool *tool)
115 {
116 struct perf_session *session = zalloc(sizeof(*session));
117
118 if (!session)
119 goto out;
120
121 session->repipe = repipe;
122 session->tool = tool;
123 machines__init(&session->machines);
124 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
125
126 if (file) {
127 if (perf_data_file__open(file))
128 goto out_delete;
129
130 session->file = file;
131
132 if (perf_data_file__is_read(file)) {
133 if (perf_session__open(session) < 0)
134 goto out_close;
135
136 perf_session__set_id_hdr_size(session);
137 perf_session__set_comm_exec(session);
138 }
139 }
140
141 if (!file || perf_data_file__is_write(file)) {
142 /*
143 * In O_RDONLY mode this will be performed when reading the
144 * kernel MMAP event, in perf_event__process_mmap().
145 */
146 if (perf_session__create_kernel_maps(session) < 0)
147 pr_warning("Cannot read kernel map\n");
148 }
149
150 if (tool && tool->ordering_requires_timestamps &&
151 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
152 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
153 tool->ordered_events = false;
154 }
155
156 return session;
157
158 out_close:
159 perf_data_file__close(file);
160 out_delete:
161 perf_session__delete(session);
162 out:
163 return NULL;
164 }
165
166 static void perf_session__delete_threads(struct perf_session *session)
167 {
168 machine__delete_threads(&session->machines.host);
169 }
170
171 static void perf_session_env__delete(struct perf_session_env *env)
172 {
173 zfree(&env->hostname);
174 zfree(&env->os_release);
175 zfree(&env->version);
176 zfree(&env->arch);
177 zfree(&env->cpu_desc);
178 zfree(&env->cpuid);
179
180 zfree(&env->cmdline);
181 zfree(&env->sibling_cores);
182 zfree(&env->sibling_threads);
183 zfree(&env->numa_nodes);
184 zfree(&env->pmu_mappings);
185 }
186
187 void perf_session__delete(struct perf_session *session)
188 {
189 auxtrace__free(session);
190 perf_session__destroy_kernel_maps(session);
191 perf_session__delete_threads(session);
192 perf_session_env__delete(&session->header.env);
193 machines__exit(&session->machines);
194 if (session->file)
195 perf_data_file__close(session->file);
196 free(session);
197 }
198
199 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
200 __maybe_unused,
201 union perf_event *event
202 __maybe_unused,
203 struct perf_session *session
204 __maybe_unused)
205 {
206 dump_printf(": unhandled!\n");
207 return 0;
208 }
209
210 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
211 union perf_event *event __maybe_unused,
212 struct perf_evlist **pevlist
213 __maybe_unused)
214 {
215 dump_printf(": unhandled!\n");
216 return 0;
217 }
218
219 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
220 union perf_event *event __maybe_unused,
221 struct perf_sample *sample __maybe_unused,
222 struct perf_evsel *evsel __maybe_unused,
223 struct machine *machine __maybe_unused)
224 {
225 dump_printf(": unhandled!\n");
226 return 0;
227 }
228
229 static int process_event_stub(struct perf_tool *tool __maybe_unused,
230 union perf_event *event __maybe_unused,
231 struct perf_sample *sample __maybe_unused,
232 struct machine *machine __maybe_unused)
233 {
234 dump_printf(": unhandled!\n");
235 return 0;
236 }
237
238 static int process_build_id_stub(struct perf_tool *tool __maybe_unused,
239 union perf_event *event __maybe_unused,
240 struct perf_session *session __maybe_unused)
241 {
242 dump_printf(": unhandled!\n");
243 return 0;
244 }
245
246 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
247 union perf_event *event __maybe_unused,
248 struct ordered_events *oe __maybe_unused)
249 {
250 dump_printf(": unhandled!\n");
251 return 0;
252 }
253
254 static int process_finished_round(struct perf_tool *tool,
255 union perf_event *event,
256 struct ordered_events *oe);
257
258 static int process_id_index_stub(struct perf_tool *tool __maybe_unused,
259 union perf_event *event __maybe_unused,
260 struct perf_session *perf_session
261 __maybe_unused)
262 {
263 dump_printf(": unhandled!\n");
264 return 0;
265 }
266
267 static int process_event_auxtrace_info_stub(struct perf_tool *tool __maybe_unused,
268 union perf_event *event __maybe_unused,
269 struct perf_session *session __maybe_unused)
270 {
271 dump_printf(": unhandled!\n");
272 return 0;
273 }
274
275 static int skipn(int fd, off_t n)
276 {
277 char buf[4096];
278 ssize_t ret;
279
280 while (n > 0) {
281 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
282 if (ret <= 0)
283 return ret;
284 n -= ret;
285 }
286
287 return 0;
288 }
289
290 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
291 union perf_event *event,
292 struct perf_session *session
293 __maybe_unused)
294 {
295 dump_printf(": unhandled!\n");
296 if (perf_data_file__is_pipe(session->file))
297 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
298 return event->auxtrace.size;
299 }
300
301 static
302 int process_event_auxtrace_error_stub(struct perf_tool *tool __maybe_unused,
303 union perf_event *event __maybe_unused,
304 struct perf_session *session __maybe_unused)
305 {
306 dump_printf(": unhandled!\n");
307 return 0;
308 }
309
310 void perf_tool__fill_defaults(struct perf_tool *tool)
311 {
312 if (tool->sample == NULL)
313 tool->sample = process_event_sample_stub;
314 if (tool->mmap == NULL)
315 tool->mmap = process_event_stub;
316 if (tool->mmap2 == NULL)
317 tool->mmap2 = process_event_stub;
318 if (tool->comm == NULL)
319 tool->comm = process_event_stub;
320 if (tool->fork == NULL)
321 tool->fork = process_event_stub;
322 if (tool->exit == NULL)
323 tool->exit = process_event_stub;
324 if (tool->lost == NULL)
325 tool->lost = perf_event__process_lost;
326 if (tool->read == NULL)
327 tool->read = process_event_sample_stub;
328 if (tool->throttle == NULL)
329 tool->throttle = process_event_stub;
330 if (tool->unthrottle == NULL)
331 tool->unthrottle = process_event_stub;
332 if (tool->attr == NULL)
333 tool->attr = process_event_synth_attr_stub;
334 if (tool->tracing_data == NULL)
335 tool->tracing_data = process_event_synth_tracing_data_stub;
336 if (tool->build_id == NULL)
337 tool->build_id = process_build_id_stub;
338 if (tool->finished_round == NULL) {
339 if (tool->ordered_events)
340 tool->finished_round = process_finished_round;
341 else
342 tool->finished_round = process_finished_round_stub;
343 }
344 if (tool->id_index == NULL)
345 tool->id_index = process_id_index_stub;
346 if (tool->auxtrace_info == NULL)
347 tool->auxtrace_info = process_event_auxtrace_info_stub;
348 if (tool->auxtrace == NULL)
349 tool->auxtrace = process_event_auxtrace_stub;
350 if (tool->auxtrace_error == NULL)
351 tool->auxtrace_error = process_event_auxtrace_error_stub;
352 }
353
354 static void swap_sample_id_all(union perf_event *event, void *data)
355 {
356 void *end = (void *) event + event->header.size;
357 int size = end - data;
358
359 BUG_ON(size % sizeof(u64));
360 mem_bswap_64(data, size);
361 }
362
363 static void perf_event__all64_swap(union perf_event *event,
364 bool sample_id_all __maybe_unused)
365 {
366 struct perf_event_header *hdr = &event->header;
367 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
368 }
369
370 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
371 {
372 event->comm.pid = bswap_32(event->comm.pid);
373 event->comm.tid = bswap_32(event->comm.tid);
374
375 if (sample_id_all) {
376 void *data = &event->comm.comm;
377
378 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
379 swap_sample_id_all(event, data);
380 }
381 }
382
383 static void perf_event__mmap_swap(union perf_event *event,
384 bool sample_id_all)
385 {
386 event->mmap.pid = bswap_32(event->mmap.pid);
387 event->mmap.tid = bswap_32(event->mmap.tid);
388 event->mmap.start = bswap_64(event->mmap.start);
389 event->mmap.len = bswap_64(event->mmap.len);
390 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
391
392 if (sample_id_all) {
393 void *data = &event->mmap.filename;
394
395 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
396 swap_sample_id_all(event, data);
397 }
398 }
399
400 static void perf_event__mmap2_swap(union perf_event *event,
401 bool sample_id_all)
402 {
403 event->mmap2.pid = bswap_32(event->mmap2.pid);
404 event->mmap2.tid = bswap_32(event->mmap2.tid);
405 event->mmap2.start = bswap_64(event->mmap2.start);
406 event->mmap2.len = bswap_64(event->mmap2.len);
407 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
408 event->mmap2.maj = bswap_32(event->mmap2.maj);
409 event->mmap2.min = bswap_32(event->mmap2.min);
410 event->mmap2.ino = bswap_64(event->mmap2.ino);
411
412 if (sample_id_all) {
413 void *data = &event->mmap2.filename;
414
415 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
416 swap_sample_id_all(event, data);
417 }
418 }
419 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
420 {
421 event->fork.pid = bswap_32(event->fork.pid);
422 event->fork.tid = bswap_32(event->fork.tid);
423 event->fork.ppid = bswap_32(event->fork.ppid);
424 event->fork.ptid = bswap_32(event->fork.ptid);
425 event->fork.time = bswap_64(event->fork.time);
426
427 if (sample_id_all)
428 swap_sample_id_all(event, &event->fork + 1);
429 }
430
431 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
432 {
433 event->read.pid = bswap_32(event->read.pid);
434 event->read.tid = bswap_32(event->read.tid);
435 event->read.value = bswap_64(event->read.value);
436 event->read.time_enabled = bswap_64(event->read.time_enabled);
437 event->read.time_running = bswap_64(event->read.time_running);
438 event->read.id = bswap_64(event->read.id);
439
440 if (sample_id_all)
441 swap_sample_id_all(event, &event->read + 1);
442 }
443
444 static void perf_event__throttle_swap(union perf_event *event,
445 bool sample_id_all)
446 {
447 event->throttle.time = bswap_64(event->throttle.time);
448 event->throttle.id = bswap_64(event->throttle.id);
449 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
450
451 if (sample_id_all)
452 swap_sample_id_all(event, &event->throttle + 1);
453 }
454
455 static u8 revbyte(u8 b)
456 {
457 int rev = (b >> 4) | ((b & 0xf) << 4);
458 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
459 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
460 return (u8) rev;
461 }
462
463 /*
464 * XXX this is hack in attempt to carry flags bitfield
465 * throught endian village. ABI says:
466 *
467 * Bit-fields are allocated from right to left (least to most significant)
468 * on little-endian implementations and from left to right (most to least
469 * significant) on big-endian implementations.
470 *
471 * The above seems to be byte specific, so we need to reverse each
472 * byte of the bitfield. 'Internet' also says this might be implementation
473 * specific and we probably need proper fix and carry perf_event_attr
474 * bitfield flags in separate data file FEAT_ section. Thought this seems
475 * to work for now.
476 */
477 static void swap_bitfield(u8 *p, unsigned len)
478 {
479 unsigned i;
480
481 for (i = 0; i < len; i++) {
482 *p = revbyte(*p);
483 p++;
484 }
485 }
486
487 /* exported for swapping attributes in file header */
488 void perf_event__attr_swap(struct perf_event_attr *attr)
489 {
490 attr->type = bswap_32(attr->type);
491 attr->size = bswap_32(attr->size);
492 attr->config = bswap_64(attr->config);
493 attr->sample_period = bswap_64(attr->sample_period);
494 attr->sample_type = bswap_64(attr->sample_type);
495 attr->read_format = bswap_64(attr->read_format);
496 attr->wakeup_events = bswap_32(attr->wakeup_events);
497 attr->bp_type = bswap_32(attr->bp_type);
498 attr->bp_addr = bswap_64(attr->bp_addr);
499 attr->bp_len = bswap_64(attr->bp_len);
500 attr->branch_sample_type = bswap_64(attr->branch_sample_type);
501 attr->sample_regs_user = bswap_64(attr->sample_regs_user);
502 attr->sample_stack_user = bswap_32(attr->sample_stack_user);
503 attr->aux_watermark = bswap_32(attr->aux_watermark);
504
505 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64));
506 }
507
508 static void perf_event__hdr_attr_swap(union perf_event *event,
509 bool sample_id_all __maybe_unused)
510 {
511 size_t size;
512
513 perf_event__attr_swap(&event->attr.attr);
514
515 size = event->header.size;
516 size -= (void *)&event->attr.id - (void *)event;
517 mem_bswap_64(event->attr.id, size);
518 }
519
520 static void perf_event__event_type_swap(union perf_event *event,
521 bool sample_id_all __maybe_unused)
522 {
523 event->event_type.event_type.event_id =
524 bswap_64(event->event_type.event_type.event_id);
525 }
526
527 static void perf_event__tracing_data_swap(union perf_event *event,
528 bool sample_id_all __maybe_unused)
529 {
530 event->tracing_data.size = bswap_32(event->tracing_data.size);
531 }
532
533 static void perf_event__auxtrace_info_swap(union perf_event *event,
534 bool sample_id_all __maybe_unused)
535 {
536 size_t size;
537
538 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
539
540 size = event->header.size;
541 size -= (void *)&event->auxtrace_info.priv - (void *)event;
542 mem_bswap_64(event->auxtrace_info.priv, size);
543 }
544
545 static void perf_event__auxtrace_swap(union perf_event *event,
546 bool sample_id_all __maybe_unused)
547 {
548 event->auxtrace.size = bswap_64(event->auxtrace.size);
549 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
550 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
551 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
552 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
553 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
554 }
555
556 static void perf_event__auxtrace_error_swap(union perf_event *event,
557 bool sample_id_all __maybe_unused)
558 {
559 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
560 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
561 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
562 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
563 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
564 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
565 }
566
567 typedef void (*perf_event__swap_op)(union perf_event *event,
568 bool sample_id_all);
569
570 static perf_event__swap_op perf_event__swap_ops[] = {
571 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
572 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
573 [PERF_RECORD_COMM] = perf_event__comm_swap,
574 [PERF_RECORD_FORK] = perf_event__task_swap,
575 [PERF_RECORD_EXIT] = perf_event__task_swap,
576 [PERF_RECORD_LOST] = perf_event__all64_swap,
577 [PERF_RECORD_READ] = perf_event__read_swap,
578 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
579 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
580 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
581 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
582 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
583 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
584 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
585 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
586 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
587 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
588 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
589 [PERF_RECORD_HEADER_MAX] = NULL,
590 };
591
592 /*
593 * When perf record finishes a pass on every buffers, it records this pseudo
594 * event.
595 * We record the max timestamp t found in the pass n.
596 * Assuming these timestamps are monotonic across cpus, we know that if
597 * a buffer still has events with timestamps below t, they will be all
598 * available and then read in the pass n + 1.
599 * Hence when we start to read the pass n + 2, we can safely flush every
600 * events with timestamps below t.
601 *
602 * ============ PASS n =================
603 * CPU 0 | CPU 1
604 * |
605 * cnt1 timestamps | cnt2 timestamps
606 * 1 | 2
607 * 2 | 3
608 * - | 4 <--- max recorded
609 *
610 * ============ PASS n + 1 ==============
611 * CPU 0 | CPU 1
612 * |
613 * cnt1 timestamps | cnt2 timestamps
614 * 3 | 5
615 * 4 | 6
616 * 5 | 7 <---- max recorded
617 *
618 * Flush every events below timestamp 4
619 *
620 * ============ PASS n + 2 ==============
621 * CPU 0 | CPU 1
622 * |
623 * cnt1 timestamps | cnt2 timestamps
624 * 6 | 8
625 * 7 | 9
626 * - | 10
627 *
628 * Flush every events below timestamp 7
629 * etc...
630 */
631 static int process_finished_round(struct perf_tool *tool __maybe_unused,
632 union perf_event *event __maybe_unused,
633 struct ordered_events *oe)
634 {
635 return ordered_events__flush(oe, OE_FLUSH__ROUND);
636 }
637
638 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
639 struct perf_sample *sample, u64 file_offset)
640 {
641 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
642 }
643
644 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
645 {
646 struct ip_callchain *callchain = sample->callchain;
647 struct branch_stack *lbr_stack = sample->branch_stack;
648 u64 kernel_callchain_nr = callchain->nr;
649 unsigned int i;
650
651 for (i = 0; i < kernel_callchain_nr; i++) {
652 if (callchain->ips[i] == PERF_CONTEXT_USER)
653 break;
654 }
655
656 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
657 u64 total_nr;
658 /*
659 * LBR callstack can only get user call chain,
660 * i is kernel call chain number,
661 * 1 is PERF_CONTEXT_USER.
662 *
663 * The user call chain is stored in LBR registers.
664 * LBR are pair registers. The caller is stored
665 * in "from" register, while the callee is stored
666 * in "to" register.
667 * For example, there is a call stack
668 * "A"->"B"->"C"->"D".
669 * The LBR registers will recorde like
670 * "C"->"D", "B"->"C", "A"->"B".
671 * So only the first "to" register and all "from"
672 * registers are needed to construct the whole stack.
673 */
674 total_nr = i + 1 + lbr_stack->nr + 1;
675 kernel_callchain_nr = i + 1;
676
677 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
678
679 for (i = 0; i < kernel_callchain_nr; i++)
680 printf("..... %2d: %016" PRIx64 "\n",
681 i, callchain->ips[i]);
682
683 printf("..... %2d: %016" PRIx64 "\n",
684 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
685 for (i = 0; i < lbr_stack->nr; i++)
686 printf("..... %2d: %016" PRIx64 "\n",
687 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
688 }
689 }
690
691 static void callchain__printf(struct perf_evsel *evsel,
692 struct perf_sample *sample)
693 {
694 unsigned int i;
695 struct ip_callchain *callchain = sample->callchain;
696
697 if (has_branch_callstack(evsel))
698 callchain__lbr_callstack_printf(sample);
699
700 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
701
702 for (i = 0; i < callchain->nr; i++)
703 printf("..... %2d: %016" PRIx64 "\n",
704 i, callchain->ips[i]);
705 }
706
707 static void branch_stack__printf(struct perf_sample *sample)
708 {
709 uint64_t i;
710
711 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
712
713 for (i = 0; i < sample->branch_stack->nr; i++)
714 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n",
715 i, sample->branch_stack->entries[i].from,
716 sample->branch_stack->entries[i].to);
717 }
718
719 static void regs_dump__printf(u64 mask, u64 *regs)
720 {
721 unsigned rid, i = 0;
722
723 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
724 u64 val = regs[i++];
725
726 printf(".... %-5s 0x%" PRIx64 "\n",
727 perf_reg_name(rid), val);
728 }
729 }
730
731 static const char *regs_abi[] = {
732 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
733 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
734 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
735 };
736
737 static inline const char *regs_dump_abi(struct regs_dump *d)
738 {
739 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
740 return "unknown";
741
742 return regs_abi[d->abi];
743 }
744
745 static void regs__printf(const char *type, struct regs_dump *regs)
746 {
747 u64 mask = regs->mask;
748
749 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
750 type,
751 mask,
752 regs_dump_abi(regs));
753
754 regs_dump__printf(mask, regs->regs);
755 }
756
757 static void regs_user__printf(struct perf_sample *sample)
758 {
759 struct regs_dump *user_regs = &sample->user_regs;
760
761 if (user_regs->regs)
762 regs__printf("user", user_regs);
763 }
764
765 static void regs_intr__printf(struct perf_sample *sample)
766 {
767 struct regs_dump *intr_regs = &sample->intr_regs;
768
769 if (intr_regs->regs)
770 regs__printf("intr", intr_regs);
771 }
772
773 static void stack_user__printf(struct stack_dump *dump)
774 {
775 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
776 dump->size, dump->offset);
777 }
778
779 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
780 union perf_event *event,
781 struct perf_sample *sample)
782 {
783 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
784
785 if (event->header.type != PERF_RECORD_SAMPLE &&
786 !perf_evlist__sample_id_all(evlist)) {
787 fputs("-1 -1 ", stdout);
788 return;
789 }
790
791 if ((sample_type & PERF_SAMPLE_CPU))
792 printf("%u ", sample->cpu);
793
794 if (sample_type & PERF_SAMPLE_TIME)
795 printf("%" PRIu64 " ", sample->time);
796 }
797
798 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
799 {
800 printf("... sample_read:\n");
801
802 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
803 printf("...... time enabled %016" PRIx64 "\n",
804 sample->read.time_enabled);
805
806 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
807 printf("...... time running %016" PRIx64 "\n",
808 sample->read.time_running);
809
810 if (read_format & PERF_FORMAT_GROUP) {
811 u64 i;
812
813 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
814
815 for (i = 0; i < sample->read.group.nr; i++) {
816 struct sample_read_value *value;
817
818 value = &sample->read.group.values[i];
819 printf("..... id %016" PRIx64
820 ", value %016" PRIx64 "\n",
821 value->id, value->value);
822 }
823 } else
824 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
825 sample->read.one.id, sample->read.one.value);
826 }
827
828 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
829 u64 file_offset, struct perf_sample *sample)
830 {
831 if (!dump_trace)
832 return;
833
834 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
835 file_offset, event->header.size, event->header.type);
836
837 trace_event(event);
838
839 if (sample)
840 perf_evlist__print_tstamp(evlist, event, sample);
841
842 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
843 event->header.size, perf_event__name(event->header.type));
844 }
845
846 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
847 struct perf_sample *sample)
848 {
849 u64 sample_type;
850
851 if (!dump_trace)
852 return;
853
854 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
855 event->header.misc, sample->pid, sample->tid, sample->ip,
856 sample->period, sample->addr);
857
858 sample_type = evsel->attr.sample_type;
859
860 if (sample_type & PERF_SAMPLE_CALLCHAIN)
861 callchain__printf(evsel, sample);
862
863 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel))
864 branch_stack__printf(sample);
865
866 if (sample_type & PERF_SAMPLE_REGS_USER)
867 regs_user__printf(sample);
868
869 if (sample_type & PERF_SAMPLE_REGS_INTR)
870 regs_intr__printf(sample);
871
872 if (sample_type & PERF_SAMPLE_STACK_USER)
873 stack_user__printf(&sample->user_stack);
874
875 if (sample_type & PERF_SAMPLE_WEIGHT)
876 printf("... weight: %" PRIu64 "\n", sample->weight);
877
878 if (sample_type & PERF_SAMPLE_DATA_SRC)
879 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
880
881 if (sample_type & PERF_SAMPLE_TRANSACTION)
882 printf("... transaction: %" PRIx64 "\n", sample->transaction);
883
884 if (sample_type & PERF_SAMPLE_READ)
885 sample_read__printf(sample, evsel->attr.read_format);
886 }
887
888 static struct machine *machines__find_for_cpumode(struct machines *machines,
889 union perf_event *event,
890 struct perf_sample *sample)
891 {
892 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
893 struct machine *machine;
894
895 if (perf_guest &&
896 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
897 (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
898 u32 pid;
899
900 if (event->header.type == PERF_RECORD_MMAP
901 || event->header.type == PERF_RECORD_MMAP2)
902 pid = event->mmap.pid;
903 else
904 pid = sample->pid;
905
906 machine = machines__find(machines, pid);
907 if (!machine)
908 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID);
909 return machine;
910 }
911
912 return &machines->host;
913 }
914
915 static int deliver_sample_value(struct perf_evlist *evlist,
916 struct perf_tool *tool,
917 union perf_event *event,
918 struct perf_sample *sample,
919 struct sample_read_value *v,
920 struct machine *machine)
921 {
922 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
923
924 if (sid) {
925 sample->id = v->id;
926 sample->period = v->value - sid->period;
927 sid->period = v->value;
928 }
929
930 if (!sid || sid->evsel == NULL) {
931 ++evlist->stats.nr_unknown_id;
932 return 0;
933 }
934
935 return tool->sample(tool, event, sample, sid->evsel, machine);
936 }
937
938 static int deliver_sample_group(struct perf_evlist *evlist,
939 struct perf_tool *tool,
940 union perf_event *event,
941 struct perf_sample *sample,
942 struct machine *machine)
943 {
944 int ret = -EINVAL;
945 u64 i;
946
947 for (i = 0; i < sample->read.group.nr; i++) {
948 ret = deliver_sample_value(evlist, tool, event, sample,
949 &sample->read.group.values[i],
950 machine);
951 if (ret)
952 break;
953 }
954
955 return ret;
956 }
957
958 static int
959 perf_evlist__deliver_sample(struct perf_evlist *evlist,
960 struct perf_tool *tool,
961 union perf_event *event,
962 struct perf_sample *sample,
963 struct perf_evsel *evsel,
964 struct machine *machine)
965 {
966 /* We know evsel != NULL. */
967 u64 sample_type = evsel->attr.sample_type;
968 u64 read_format = evsel->attr.read_format;
969
970 /* Standard sample delievery. */
971 if (!(sample_type & PERF_SAMPLE_READ))
972 return tool->sample(tool, event, sample, evsel, machine);
973
974 /* For PERF_SAMPLE_READ we have either single or group mode. */
975 if (read_format & PERF_FORMAT_GROUP)
976 return deliver_sample_group(evlist, tool, event, sample,
977 machine);
978 else
979 return deliver_sample_value(evlist, tool, event, sample,
980 &sample->read.one, machine);
981 }
982
983 static int machines__deliver_event(struct machines *machines,
984 struct perf_evlist *evlist,
985 union perf_event *event,
986 struct perf_sample *sample,
987 struct perf_tool *tool, u64 file_offset)
988 {
989 struct perf_evsel *evsel;
990 struct machine *machine;
991
992 dump_event(evlist, event, file_offset, sample);
993
994 evsel = perf_evlist__id2evsel(evlist, sample->id);
995
996 machine = machines__find_for_cpumode(machines, event, sample);
997
998 switch (event->header.type) {
999 case PERF_RECORD_SAMPLE:
1000 dump_sample(evsel, event, sample);
1001 if (evsel == NULL) {
1002 ++evlist->stats.nr_unknown_id;
1003 return 0;
1004 }
1005 if (machine == NULL) {
1006 ++evlist->stats.nr_unprocessable_samples;
1007 return 0;
1008 }
1009 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1010 case PERF_RECORD_MMAP:
1011 return tool->mmap(tool, event, sample, machine);
1012 case PERF_RECORD_MMAP2:
1013 return tool->mmap2(tool, event, sample, machine);
1014 case PERF_RECORD_COMM:
1015 return tool->comm(tool, event, sample, machine);
1016 case PERF_RECORD_FORK:
1017 return tool->fork(tool, event, sample, machine);
1018 case PERF_RECORD_EXIT:
1019 return tool->exit(tool, event, sample, machine);
1020 case PERF_RECORD_LOST:
1021 if (tool->lost == perf_event__process_lost)
1022 evlist->stats.total_lost += event->lost.lost;
1023 return tool->lost(tool, event, sample, machine);
1024 case PERF_RECORD_READ:
1025 return tool->read(tool, event, sample, evsel, machine);
1026 case PERF_RECORD_THROTTLE:
1027 return tool->throttle(tool, event, sample, machine);
1028 case PERF_RECORD_UNTHROTTLE:
1029 return tool->unthrottle(tool, event, sample, machine);
1030 default:
1031 ++evlist->stats.nr_unknown_events;
1032 return -1;
1033 }
1034 }
1035
1036 static int perf_session__deliver_event(struct perf_session *session,
1037 union perf_event *event,
1038 struct perf_sample *sample,
1039 struct perf_tool *tool,
1040 u64 file_offset)
1041 {
1042 int ret;
1043
1044 ret = auxtrace__process_event(session, event, sample, tool);
1045 if (ret < 0)
1046 return ret;
1047 if (ret > 0)
1048 return 0;
1049
1050 return machines__deliver_event(&session->machines, session->evlist,
1051 event, sample, tool, file_offset);
1052 }
1053
1054 static s64 perf_session__process_user_event(struct perf_session *session,
1055 union perf_event *event,
1056 u64 file_offset)
1057 {
1058 struct ordered_events *oe = &session->ordered_events;
1059 struct perf_tool *tool = session->tool;
1060 int fd = perf_data_file__fd(session->file);
1061 int err;
1062
1063 dump_event(session->evlist, event, file_offset, NULL);
1064
1065 /* These events are processed right away */
1066 switch (event->header.type) {
1067 case PERF_RECORD_HEADER_ATTR:
1068 err = tool->attr(tool, event, &session->evlist);
1069 if (err == 0) {
1070 perf_session__set_id_hdr_size(session);
1071 perf_session__set_comm_exec(session);
1072 }
1073 return err;
1074 case PERF_RECORD_HEADER_EVENT_TYPE:
1075 /*
1076 * Depreceated, but we need to handle it for sake
1077 * of old data files create in pipe mode.
1078 */
1079 return 0;
1080 case PERF_RECORD_HEADER_TRACING_DATA:
1081 /* setup for reading amidst mmap */
1082 lseek(fd, file_offset, SEEK_SET);
1083 return tool->tracing_data(tool, event, session);
1084 case PERF_RECORD_HEADER_BUILD_ID:
1085 return tool->build_id(tool, event, session);
1086 case PERF_RECORD_FINISHED_ROUND:
1087 return tool->finished_round(tool, event, oe);
1088 case PERF_RECORD_ID_INDEX:
1089 return tool->id_index(tool, event, session);
1090 case PERF_RECORD_AUXTRACE_INFO:
1091 return tool->auxtrace_info(tool, event, session);
1092 case PERF_RECORD_AUXTRACE:
1093 /* setup for reading amidst mmap */
1094 lseek(fd, file_offset + event->header.size, SEEK_SET);
1095 return tool->auxtrace(tool, event, session);
1096 case PERF_RECORD_AUXTRACE_ERROR:
1097 perf_session__auxtrace_error_inc(session, event);
1098 return tool->auxtrace_error(tool, event, session);
1099 default:
1100 return -EINVAL;
1101 }
1102 }
1103
1104 int perf_session__deliver_synth_event(struct perf_session *session,
1105 union perf_event *event,
1106 struct perf_sample *sample)
1107 {
1108 struct perf_evlist *evlist = session->evlist;
1109 struct perf_tool *tool = session->tool;
1110
1111 events_stats__inc(&evlist->stats, event->header.type);
1112
1113 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1114 return perf_session__process_user_event(session, event, 0);
1115
1116 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1117 }
1118
1119 static void event_swap(union perf_event *event, bool sample_id_all)
1120 {
1121 perf_event__swap_op swap;
1122
1123 swap = perf_event__swap_ops[event->header.type];
1124 if (swap)
1125 swap(event, sample_id_all);
1126 }
1127
1128 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1129 void *buf, size_t buf_sz,
1130 union perf_event **event_ptr,
1131 struct perf_sample *sample)
1132 {
1133 union perf_event *event;
1134 size_t hdr_sz, rest;
1135 int fd;
1136
1137 if (session->one_mmap && !session->header.needs_swap) {
1138 event = file_offset - session->one_mmap_offset +
1139 session->one_mmap_addr;
1140 goto out_parse_sample;
1141 }
1142
1143 if (perf_data_file__is_pipe(session->file))
1144 return -1;
1145
1146 fd = perf_data_file__fd(session->file);
1147 hdr_sz = sizeof(struct perf_event_header);
1148
1149 if (buf_sz < hdr_sz)
1150 return -1;
1151
1152 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1153 readn(fd, &buf, hdr_sz) != (ssize_t)hdr_sz)
1154 return -1;
1155
1156 event = (union perf_event *)buf;
1157
1158 if (session->header.needs_swap)
1159 perf_event_header__bswap(&event->header);
1160
1161 if (event->header.size < hdr_sz)
1162 return -1;
1163
1164 rest = event->header.size - hdr_sz;
1165
1166 if (readn(fd, &buf, rest) != (ssize_t)rest)
1167 return -1;
1168
1169 if (session->header.needs_swap)
1170 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1171
1172 out_parse_sample:
1173
1174 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1175 perf_evlist__parse_sample(session->evlist, event, sample))
1176 return -1;
1177
1178 *event_ptr = event;
1179
1180 return 0;
1181 }
1182
1183 static s64 perf_session__process_event(struct perf_session *session,
1184 union perf_event *event, u64 file_offset)
1185 {
1186 struct perf_evlist *evlist = session->evlist;
1187 struct perf_tool *tool = session->tool;
1188 struct perf_sample sample;
1189 int ret;
1190
1191 if (session->header.needs_swap)
1192 event_swap(event, perf_evlist__sample_id_all(evlist));
1193
1194 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1195 return -EINVAL;
1196
1197 events_stats__inc(&evlist->stats, event->header.type);
1198
1199 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1200 return perf_session__process_user_event(session, event, file_offset);
1201
1202 /*
1203 * For all kernel events we get the sample data
1204 */
1205 ret = perf_evlist__parse_sample(evlist, event, &sample);
1206 if (ret)
1207 return ret;
1208
1209 if (tool->ordered_events) {
1210 ret = perf_session__queue_event(session, event, &sample, file_offset);
1211 if (ret != -ETIME)
1212 return ret;
1213 }
1214
1215 return perf_session__deliver_event(session, event, &sample, tool,
1216 file_offset);
1217 }
1218
1219 void perf_event_header__bswap(struct perf_event_header *hdr)
1220 {
1221 hdr->type = bswap_32(hdr->type);
1222 hdr->misc = bswap_16(hdr->misc);
1223 hdr->size = bswap_16(hdr->size);
1224 }
1225
1226 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1227 {
1228 return machine__findnew_thread(&session->machines.host, -1, pid);
1229 }
1230
1231 static struct thread *perf_session__register_idle_thread(struct perf_session *session)
1232 {
1233 struct thread *thread;
1234
1235 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1236 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1237 pr_err("problem inserting idle task.\n");
1238 thread = NULL;
1239 }
1240
1241 return thread;
1242 }
1243
1244 static void perf_session__warn_about_errors(const struct perf_session *session)
1245 {
1246 const struct events_stats *stats = &session->evlist->stats;
1247 const struct ordered_events *oe = &session->ordered_events;
1248
1249 if (session->tool->lost == perf_event__process_lost &&
1250 stats->nr_events[PERF_RECORD_LOST] != 0) {
1251 ui__warning("Processed %d events and lost %d chunks!\n\n"
1252 "Check IO/CPU overload!\n\n",
1253 stats->nr_events[0],
1254 stats->nr_events[PERF_RECORD_LOST]);
1255 }
1256
1257 if (stats->nr_unknown_events != 0) {
1258 ui__warning("Found %u unknown events!\n\n"
1259 "Is this an older tool processing a perf.data "
1260 "file generated by a more recent tool?\n\n"
1261 "If that is not the case, consider "
1262 "reporting to linux-kernel@vger.kernel.org.\n\n",
1263 stats->nr_unknown_events);
1264 }
1265
1266 if (stats->nr_unknown_id != 0) {
1267 ui__warning("%u samples with id not present in the header\n",
1268 stats->nr_unknown_id);
1269 }
1270
1271 if (stats->nr_invalid_chains != 0) {
1272 ui__warning("Found invalid callchains!\n\n"
1273 "%u out of %u events were discarded for this reason.\n\n"
1274 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1275 stats->nr_invalid_chains,
1276 stats->nr_events[PERF_RECORD_SAMPLE]);
1277 }
1278
1279 if (stats->nr_unprocessable_samples != 0) {
1280 ui__warning("%u unprocessable samples recorded.\n"
1281 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1282 stats->nr_unprocessable_samples);
1283 }
1284
1285 if (oe->nr_unordered_events != 0)
1286 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1287
1288 events_stats__auxtrace_error_warn(stats);
1289 }
1290
1291 volatile int session_done;
1292
1293 static int __perf_session__process_pipe_events(struct perf_session *session)
1294 {
1295 struct ordered_events *oe = &session->ordered_events;
1296 struct perf_tool *tool = session->tool;
1297 int fd = perf_data_file__fd(session->file);
1298 union perf_event *event;
1299 uint32_t size, cur_size = 0;
1300 void *buf = NULL;
1301 s64 skip = 0;
1302 u64 head;
1303 ssize_t err;
1304 void *p;
1305
1306 perf_tool__fill_defaults(tool);
1307
1308 head = 0;
1309 cur_size = sizeof(union perf_event);
1310
1311 buf = malloc(cur_size);
1312 if (!buf)
1313 return -errno;
1314 more:
1315 event = buf;
1316 err = readn(fd, event, sizeof(struct perf_event_header));
1317 if (err <= 0) {
1318 if (err == 0)
1319 goto done;
1320
1321 pr_err("failed to read event header\n");
1322 goto out_err;
1323 }
1324
1325 if (session->header.needs_swap)
1326 perf_event_header__bswap(&event->header);
1327
1328 size = event->header.size;
1329 if (size < sizeof(struct perf_event_header)) {
1330 pr_err("bad event header size\n");
1331 goto out_err;
1332 }
1333
1334 if (size > cur_size) {
1335 void *new = realloc(buf, size);
1336 if (!new) {
1337 pr_err("failed to allocate memory to read event\n");
1338 goto out_err;
1339 }
1340 buf = new;
1341 cur_size = size;
1342 event = buf;
1343 }
1344 p = event;
1345 p += sizeof(struct perf_event_header);
1346
1347 if (size - sizeof(struct perf_event_header)) {
1348 err = readn(fd, p, size - sizeof(struct perf_event_header));
1349 if (err <= 0) {
1350 if (err == 0) {
1351 pr_err("unexpected end of event stream\n");
1352 goto done;
1353 }
1354
1355 pr_err("failed to read event data\n");
1356 goto out_err;
1357 }
1358 }
1359
1360 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1361 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1362 head, event->header.size, event->header.type);
1363 err = -EINVAL;
1364 goto out_err;
1365 }
1366
1367 head += size;
1368
1369 if (skip > 0)
1370 head += skip;
1371
1372 if (!session_done())
1373 goto more;
1374 done:
1375 /* do the final flush for ordered samples */
1376 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1377 if (err)
1378 goto out_err;
1379 err = auxtrace__flush_events(session, tool);
1380 out_err:
1381 free(buf);
1382 perf_session__warn_about_errors(session);
1383 ordered_events__free(&session->ordered_events);
1384 auxtrace__free_events(session);
1385 return err;
1386 }
1387
1388 static union perf_event *
1389 fetch_mmaped_event(struct perf_session *session,
1390 u64 head, size_t mmap_size, char *buf)
1391 {
1392 union perf_event *event;
1393
1394 /*
1395 * Ensure we have enough space remaining to read
1396 * the size of the event in the headers.
1397 */
1398 if (head + sizeof(event->header) > mmap_size)
1399 return NULL;
1400
1401 event = (union perf_event *)(buf + head);
1402
1403 if (session->header.needs_swap)
1404 perf_event_header__bswap(&event->header);
1405
1406 if (head + event->header.size > mmap_size) {
1407 /* We're not fetching the event so swap back again */
1408 if (session->header.needs_swap)
1409 perf_event_header__bswap(&event->header);
1410 return NULL;
1411 }
1412
1413 return event;
1414 }
1415
1416 /*
1417 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1418 * slices. On 32bit we use 32MB.
1419 */
1420 #if BITS_PER_LONG == 64
1421 #define MMAP_SIZE ULLONG_MAX
1422 #define NUM_MMAPS 1
1423 #else
1424 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1425 #define NUM_MMAPS 128
1426 #endif
1427
1428 static int __perf_session__process_events(struct perf_session *session,
1429 u64 data_offset, u64 data_size,
1430 u64 file_size)
1431 {
1432 struct ordered_events *oe = &session->ordered_events;
1433 struct perf_tool *tool = session->tool;
1434 int fd = perf_data_file__fd(session->file);
1435 u64 head, page_offset, file_offset, file_pos, size;
1436 int err, mmap_prot, mmap_flags, map_idx = 0;
1437 size_t mmap_size;
1438 char *buf, *mmaps[NUM_MMAPS];
1439 union perf_event *event;
1440 struct ui_progress prog;
1441 s64 skip;
1442
1443 perf_tool__fill_defaults(tool);
1444
1445 page_offset = page_size * (data_offset / page_size);
1446 file_offset = page_offset;
1447 head = data_offset - page_offset;
1448
1449 if (data_size && (data_offset + data_size < file_size))
1450 file_size = data_offset + data_size;
1451
1452 ui_progress__init(&prog, file_size, "Processing events...");
1453
1454 mmap_size = MMAP_SIZE;
1455 if (mmap_size > file_size) {
1456 mmap_size = file_size;
1457 session->one_mmap = true;
1458 }
1459
1460 memset(mmaps, 0, sizeof(mmaps));
1461
1462 mmap_prot = PROT_READ;
1463 mmap_flags = MAP_SHARED;
1464
1465 if (session->header.needs_swap) {
1466 mmap_prot |= PROT_WRITE;
1467 mmap_flags = MAP_PRIVATE;
1468 }
1469 remap:
1470 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1471 file_offset);
1472 if (buf == MAP_FAILED) {
1473 pr_err("failed to mmap file\n");
1474 err = -errno;
1475 goto out_err;
1476 }
1477 mmaps[map_idx] = buf;
1478 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1479 file_pos = file_offset + head;
1480 if (session->one_mmap) {
1481 session->one_mmap_addr = buf;
1482 session->one_mmap_offset = file_offset;
1483 }
1484
1485 more:
1486 event = fetch_mmaped_event(session, head, mmap_size, buf);
1487 if (!event) {
1488 if (mmaps[map_idx]) {
1489 munmap(mmaps[map_idx], mmap_size);
1490 mmaps[map_idx] = NULL;
1491 }
1492
1493 page_offset = page_size * (head / page_size);
1494 file_offset += page_offset;
1495 head -= page_offset;
1496 goto remap;
1497 }
1498
1499 size = event->header.size;
1500
1501 if (size < sizeof(struct perf_event_header) ||
1502 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1503 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1504 file_offset + head, event->header.size,
1505 event->header.type);
1506 err = -EINVAL;
1507 goto out_err;
1508 }
1509
1510 if (skip)
1511 size += skip;
1512
1513 head += size;
1514 file_pos += size;
1515
1516 ui_progress__update(&prog, size);
1517
1518 if (session_done())
1519 goto out;
1520
1521 if (file_pos < file_size)
1522 goto more;
1523
1524 out:
1525 /* do the final flush for ordered samples */
1526 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1527 if (err)
1528 goto out_err;
1529 err = auxtrace__flush_events(session, tool);
1530 out_err:
1531 ui_progress__finish();
1532 perf_session__warn_about_errors(session);
1533 ordered_events__free(&session->ordered_events);
1534 auxtrace__free_events(session);
1535 session->one_mmap = false;
1536 return err;
1537 }
1538
1539 int perf_session__process_events(struct perf_session *session)
1540 {
1541 u64 size = perf_data_file__size(session->file);
1542 int err;
1543
1544 if (perf_session__register_idle_thread(session) == NULL)
1545 return -ENOMEM;
1546
1547 if (!perf_data_file__is_pipe(session->file))
1548 err = __perf_session__process_events(session,
1549 session->header.data_offset,
1550 session->header.data_size, size);
1551 else
1552 err = __perf_session__process_pipe_events(session);
1553
1554 return err;
1555 }
1556
1557 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1558 {
1559 struct perf_evsel *evsel;
1560
1561 evlist__for_each(session->evlist, evsel) {
1562 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1563 return true;
1564 }
1565
1566 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1567 return false;
1568 }
1569
1570 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1571 const char *symbol_name, u64 addr)
1572 {
1573 char *bracket;
1574 enum map_type i;
1575 struct ref_reloc_sym *ref;
1576
1577 ref = zalloc(sizeof(struct ref_reloc_sym));
1578 if (ref == NULL)
1579 return -ENOMEM;
1580
1581 ref->name = strdup(symbol_name);
1582 if (ref->name == NULL) {
1583 free(ref);
1584 return -ENOMEM;
1585 }
1586
1587 bracket = strchr(ref->name, ']');
1588 if (bracket)
1589 *bracket = '\0';
1590
1591 ref->addr = addr;
1592
1593 for (i = 0; i < MAP__NR_TYPES; ++i) {
1594 struct kmap *kmap = map__kmap(maps[i]);
1595
1596 if (!kmap)
1597 continue;
1598 kmap->ref_reloc_sym = ref;
1599 }
1600
1601 return 0;
1602 }
1603
1604 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1605 {
1606 return machines__fprintf_dsos(&session->machines, fp);
1607 }
1608
1609 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1610 bool (skip)(struct dso *dso, int parm), int parm)
1611 {
1612 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1613 }
1614
1615 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1616 {
1617 size_t ret;
1618 const char *msg = "";
1619
1620 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1621 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1622
1623 ret = fprintf(fp, "Aggregated stats:%s\n", msg);
1624
1625 ret += events_stats__fprintf(&session->evlist->stats, fp);
1626 return ret;
1627 }
1628
1629 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1630 {
1631 /*
1632 * FIXME: Here we have to actually print all the machines in this
1633 * session, not just the host...
1634 */
1635 return machine__fprintf(&session->machines.host, fp);
1636 }
1637
1638 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1639 unsigned int type)
1640 {
1641 struct perf_evsel *pos;
1642
1643 evlist__for_each(session->evlist, pos) {
1644 if (pos->attr.type == type)
1645 return pos;
1646 }
1647 return NULL;
1648 }
1649
1650 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample,
1651 struct addr_location *al,
1652 unsigned int print_opts, unsigned int stack_depth)
1653 {
1654 struct callchain_cursor_node *node;
1655 int print_ip = print_opts & PRINT_IP_OPT_IP;
1656 int print_sym = print_opts & PRINT_IP_OPT_SYM;
1657 int print_dso = print_opts & PRINT_IP_OPT_DSO;
1658 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1659 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1660 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE;
1661 char s = print_oneline ? ' ' : '\t';
1662
1663 if (symbol_conf.use_callchain && sample->callchain) {
1664 struct addr_location node_al;
1665
1666 if (thread__resolve_callchain(al->thread, evsel,
1667 sample, NULL, NULL,
1668 PERF_MAX_STACK_DEPTH) != 0) {
1669 if (verbose)
1670 error("Failed to resolve callchain. Skipping\n");
1671 return;
1672 }
1673 callchain_cursor_commit(&callchain_cursor);
1674
1675 if (print_symoffset)
1676 node_al = *al;
1677
1678 while (stack_depth) {
1679 u64 addr = 0;
1680
1681 node = callchain_cursor_current(&callchain_cursor);
1682 if (!node)
1683 break;
1684
1685 if (node->sym && node->sym->ignore)
1686 goto next;
1687
1688 if (print_ip)
1689 printf("%c%16" PRIx64, s, node->ip);
1690
1691 if (node->map)
1692 addr = node->map->map_ip(node->map, node->ip);
1693
1694 if (print_sym) {
1695 printf(" ");
1696 if (print_symoffset) {
1697 node_al.addr = addr;
1698 node_al.map = node->map;
1699 symbol__fprintf_symname_offs(node->sym, &node_al, stdout);
1700 } else
1701 symbol__fprintf_symname(node->sym, stdout);
1702 }
1703
1704 if (print_dso) {
1705 printf(" (");
1706 map__fprintf_dsoname(node->map, stdout);
1707 printf(")");
1708 }
1709
1710 if (print_srcline)
1711 map__fprintf_srcline(node->map, addr, "\n ",
1712 stdout);
1713
1714 if (!print_oneline)
1715 printf("\n");
1716
1717 stack_depth--;
1718 next:
1719 callchain_cursor_advance(&callchain_cursor);
1720 }
1721
1722 } else {
1723 if (al->sym && al->sym->ignore)
1724 return;
1725
1726 if (print_ip)
1727 printf("%16" PRIx64, sample->ip);
1728
1729 if (print_sym) {
1730 printf(" ");
1731 if (print_symoffset)
1732 symbol__fprintf_symname_offs(al->sym, al,
1733 stdout);
1734 else
1735 symbol__fprintf_symname(al->sym, stdout);
1736 }
1737
1738 if (print_dso) {
1739 printf(" (");
1740 map__fprintf_dsoname(al->map, stdout);
1741 printf(")");
1742 }
1743
1744 if (print_srcline)
1745 map__fprintf_srcline(al->map, al->addr, "\n ", stdout);
1746 }
1747 }
1748
1749 int perf_session__cpu_bitmap(struct perf_session *session,
1750 const char *cpu_list, unsigned long *cpu_bitmap)
1751 {
1752 int i, err = -1;
1753 struct cpu_map *map;
1754
1755 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1756 struct perf_evsel *evsel;
1757
1758 evsel = perf_session__find_first_evtype(session, i);
1759 if (!evsel)
1760 continue;
1761
1762 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1763 pr_err("File does not contain CPU events. "
1764 "Remove -c option to proceed.\n");
1765 return -1;
1766 }
1767 }
1768
1769 map = cpu_map__new(cpu_list);
1770 if (map == NULL) {
1771 pr_err("Invalid cpu_list\n");
1772 return -1;
1773 }
1774
1775 for (i = 0; i < map->nr; i++) {
1776 int cpu = map->map[i];
1777
1778 if (cpu >= MAX_NR_CPUS) {
1779 pr_err("Requested CPU %d too large. "
1780 "Consider raising MAX_NR_CPUS\n", cpu);
1781 goto out_delete_map;
1782 }
1783
1784 set_bit(cpu, cpu_bitmap);
1785 }
1786
1787 err = 0;
1788
1789 out_delete_map:
1790 cpu_map__delete(map);
1791 return err;
1792 }
1793
1794 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1795 bool full)
1796 {
1797 struct stat st;
1798 int fd, ret;
1799
1800 if (session == NULL || fp == NULL)
1801 return;
1802
1803 fd = perf_data_file__fd(session->file);
1804
1805 ret = fstat(fd, &st);
1806 if (ret == -1)
1807 return;
1808
1809 fprintf(fp, "# ========\n");
1810 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1811 perf_header__fprintf_info(session, fp, full);
1812 fprintf(fp, "# ========\n#\n");
1813 }
1814
1815
1816 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1817 const struct perf_evsel_str_handler *assocs,
1818 size_t nr_assocs)
1819 {
1820 struct perf_evsel *evsel;
1821 size_t i;
1822 int err;
1823
1824 for (i = 0; i < nr_assocs; i++) {
1825 /*
1826 * Adding a handler for an event not in the session,
1827 * just ignore it.
1828 */
1829 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1830 if (evsel == NULL)
1831 continue;
1832
1833 err = -EEXIST;
1834 if (evsel->handler != NULL)
1835 goto out;
1836 evsel->handler = assocs[i].handler;
1837 }
1838
1839 err = 0;
1840 out:
1841 return err;
1842 }
1843
1844 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
1845 union perf_event *event,
1846 struct perf_session *session)
1847 {
1848 struct perf_evlist *evlist = session->evlist;
1849 struct id_index_event *ie = &event->id_index;
1850 size_t i, nr, max_nr;
1851
1852 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
1853 sizeof(struct id_index_entry);
1854 nr = ie->nr;
1855 if (nr > max_nr)
1856 return -EINVAL;
1857
1858 if (dump_trace)
1859 fprintf(stdout, " nr: %zu\n", nr);
1860
1861 for (i = 0; i < nr; i++) {
1862 struct id_index_entry *e = &ie->entries[i];
1863 struct perf_sample_id *sid;
1864
1865 if (dump_trace) {
1866 fprintf(stdout, " ... id: %"PRIu64, e->id);
1867 fprintf(stdout, " idx: %"PRIu64, e->idx);
1868 fprintf(stdout, " cpu: %"PRId64, e->cpu);
1869 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
1870 }
1871
1872 sid = perf_evlist__id2sid(evlist, e->id);
1873 if (!sid)
1874 return -ENOENT;
1875 sid->idx = e->idx;
1876 sid->cpu = e->cpu;
1877 sid->tid = e->tid;
1878 }
1879 return 0;
1880 }
1881
1882 int perf_event__synthesize_id_index(struct perf_tool *tool,
1883 perf_event__handler_t process,
1884 struct perf_evlist *evlist,
1885 struct machine *machine)
1886 {
1887 union perf_event *ev;
1888 struct perf_evsel *evsel;
1889 size_t nr = 0, i = 0, sz, max_nr, n;
1890 int err;
1891
1892 pr_debug2("Synthesizing id index\n");
1893
1894 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
1895 sizeof(struct id_index_entry);
1896
1897 evlist__for_each(evlist, evsel)
1898 nr += evsel->ids;
1899
1900 n = nr > max_nr ? max_nr : nr;
1901 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
1902 ev = zalloc(sz);
1903 if (!ev)
1904 return -ENOMEM;
1905
1906 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
1907 ev->id_index.header.size = sz;
1908 ev->id_index.nr = n;
1909
1910 evlist__for_each(evlist, evsel) {
1911 u32 j;
1912
1913 for (j = 0; j < evsel->ids; j++) {
1914 struct id_index_entry *e;
1915 struct perf_sample_id *sid;
1916
1917 if (i >= n) {
1918 err = process(tool, ev, NULL, machine);
1919 if (err)
1920 goto out_err;
1921 nr -= n;
1922 i = 0;
1923 }
1924
1925 e = &ev->id_index.entries[i++];
1926
1927 e->id = evsel->id[j];
1928
1929 sid = perf_evlist__id2sid(evlist, e->id);
1930 if (!sid) {
1931 free(ev);
1932 return -ENOENT;
1933 }
1934
1935 e->idx = sid->idx;
1936 e->cpu = sid->cpu;
1937 e->tid = sid->tid;
1938 }
1939 }
1940
1941 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
1942 ev->id_index.header.size = sz;
1943 ev->id_index.nr = nr;
1944
1945 err = process(tool, ev, NULL, machine);
1946 out_err:
1947 free(ev);
1948
1949 return err;
1950 }
Linux kernel - Deliverables
This page took 0.159635 seconds and 5 git commands to generate.